National Institute of Informatics

Photometric stereo (PS) aims to recover high-fidelity surface normals by observing pixel-wise radiometric variations under different light directions. However, traditional PS methods require dense sampling of the incident light to mitigate non-Lambertian effects, such as cast shadows and specular highlights, creating a significant efficiency bottleneck for practical optical metrology. To address this efficiency bottleneck, illumination planning methods seek to identify an optimal set of light directions to maximize information gain with minimal measurements. A critical limitation of existing illumination planning paradigms is their reliance on selecting from a predefined, discrete set of candidate light directions. This discretization of the light space introduces an artificial bottleneck, severely limiting precision and adaptability. In this paper, we address this limitation by introducing a Continuous and Online ILlumination Planning framework for Photometric Stereo (COIL-PS). Instead of selecting from a fixed grid, our method formulates illumination planning as a continuous regression problem, adaptively steering light positioning in the continuous hemispherical domain. By coupling online illumination planning with feedback from intermediate normal estimates, COIL-PS adaptively navigates non-Lambertian effects such as shadows and specularities, which allow for the precise angular placement of illumination required to resolve geometric ambiguities that fall between fixed grid points. Extensive experiments on a synthetic dataset, semi-real benchmarks, and our custom-built real-world robotic validation system demonstrate that COIL-PS achieves superior normal reconstruction accuracy compared to state-of-the-art discrete planning methods, even with a budget of only ten lights, significantly outperforming discrete planning paradigms.

In this work, we measure the infrared optical constants (n, κ) of topological insulator BiSb thin films using the reflectance-transmittance method. Using the obtained (n, κ), we reveal that Bi, Bi0.9Sb0.1, and Bi0.85Sb0.15 show absorption of about 30% for film thicknesses around 20 nm and a large refractive index ranging from 6 to 9 in the infrared region. Our results suggest that BiSb has strong potential as a material for MIR detectors.

Formal reasoning about inductively defined relations and structures is widely recognized not only for its mathematical interest but also for its importance in computer science, and has applications in verifying properties of programs and algorithms. Recently, several proof systems of inductively defined predicates based on sequent calculus including the cyclic proof system CLKID-omega and the infinite-descent proof system LKID-omega have attracted much attention. Although the relation among their provabilities has been clarified so far, the logical complexity of these systems has not been much studied. The infinite-descent proof system LKID-omega is an infinite proof system for inductive definitions and allows infinite paths in proof figures. It serves as a basis for the cyclic proof system. This paper shows that the logical complexity of the provability in LKID-omega is (Pi-1-1)-complete. To show this, first it is shown that the validity for inductive definitions in standard models is equivalent to the validity for inductive definitions in standard term models. Next, using this equivalence, this paper extends the truth predicate of omega-languages, as given in Girard's textbook, to inductive definitions by employing arithmetical coding of inductive definitions. This shows that the validity of inductive definitions in standard models is a (Pi-1-1) relation. Then, using the completeness of LKID-omega for standard models, it is shown that the logical complexity of the provability in LKID-omega is (Pi-1-1)-complete.

Accurate location of Single Phase-to-Ground Faults (SPGFs) in Resonantly Grounded Networks is crucial for timely fault clearing, requiring precise dual-time detection of fault initiation and duration. However, existing methods struggle to address the weak and nonlinear characteristics of fault currents. To tackle these challenges, we propose a paradigm that integrates neural networks with enhanced Digital Signal Processing (DSP) method. This approach replaces traditional feature extraction with DSP-based methods and dynamically optimizes DSP parameters through network feedback. Specifically, we introduce a transient weak feature-enhanced attention mechanism and a learnable nonlinear waveform reconstruction mechanism, culminating in a parameter-driven DSP-enhanced Transformer named 2TD-former. Furthermore, we construct a dual-time SPGF detection dataset and publicly release it to facilitate model evaluation and support broader research. Experimental results demonstrate that our method achieves maximum errors of less than 1.4 ms for fault initiation detection and 2.8 ms for duration detection. Compared to existing fault detection methods and the state-of-the-art time series anomaly detection models, 2TDformer reduces detection errors by 30%, setting a new benchmark in performance.

Collagen fibers are significant constituents of connective tissues such as the skin, bone, and tendons. Their morphology and alignment are essential for understanding tissue function and pathology. Collagen fibers are conventionally observed using optical and electron microscopy, and Second-harmonic generation microscopy. This study introduces a novel microscopic imaging technique, Scattering Angle-Resolved Bioimaging (SARB), which utilizes the scattering properties of unstained specimens. The SARB is based on a conventional optical microscope with a structured illumination module that projects a checker pattern onto the sample. This allows the acquisition of images separated by the scattering angle, enabling the visualization of the scattering characteristics. We applied SARB and electron microscopy to thin, unstained formalin-fixed, paraffin-embedded (FFPE) sections of mouse skin. SARB successfully identified the location of collagen fiber bundles and revealed the scattering information within them. Importantly, these scattering patterns correlated with the orientation of the collagen fibrils, as observed in Transmission Electron Microscopy images. The use of unstained FFPE sections simplifies sample preparation, offering a practical advantage. Additionally, SARB provides access to fine structural information that is often difficult to obtain using conventional optical microscopy. These findings suggest that SARB has strong potential as a label-free, morphology-sensitive imaging method for evaluating fibrosis.

We introduce a novel urban visualization system that integrates 3D urban model (CityGML) and 360$^{\circ }$∘ walkthrough videos. By aligning the videos with the model and dynamically projecting relevant video frames onto the geometries, our system creates photorealistic urban visualizations, allowing users to intuitively interpret geospatial data from a pedestrian view.

As artificial intelligence (AI) increasingly participates in creative processes, designing effective human-AI collaboration is crucial. This study addresses a fundamental question: Should an AI partner prioritize deepening existing ideas (exploitation) or diversifying the creative space (exploration)? We investigated this through a controlled experiment with 148 participants, comparing two AI strategies based on March’s exploration-exploitation framework. Using a turn-based brainstorming system on the topic of “how to increase café sales,” we measured each AI strategy’s impact on human trust and idea adoption. Contrary to traditional creativity research that emphasizes divergence, our findings show that the convergent deepening strategy significantly outperformed the diversification approach in both building user trust and encouraging idea adoption. We found that the AI’s predictable behavior in the deepening condition was more easily understood by participants, leading to more effective collaboration. These results suggest that for effective human-AI co-creation, AI should act as a supportive, rather than competitive, partner that incrementally develops human-initiated concepts. This approach builds trust and increases the integration of AI’s ideas. Our work thus contributes to the understanding of how AI can transition from tool-centric approaches to more collaborative partnerships that potentially enhance human creativity.

Federated learning (FL) in vehicular networks faces significant challenges due to client nodes’ limited resources and highly dynamic network conditions, which hider active participation and degrade model performance. This paper proposes a novel incentive mechanism that explicitly addresses these challenges by combining a fuzzy logic-based contribution estimation algorithm with a reinforcement learning-based rewarding module. The estimation algorithm evaluates each client’s potential contribution by considering mobility patterns, link quality, and computational capacity, while the rewarding module dynamically adjusts incentives based on both estimated potential and actual performance. The estimation algorithm evaluates each client’s potential contribution by considering mobility patterns, link quality, and computational capacity, while the rewarding module dynamically adjusts incentives based on both estimated potential and actual performance. Extensive simulations on realistic vehicular network scenarios show that our scheme increases client participation by approximately 37% and shortens the average convergence time by about 30 global epoches compared with traditional incentive approaches. These results demonstrate that our method effectively enhances cooperation and model performance in resource-constrained, dynamic vehicular FL environments.

This paper explores a near-field integrated sensing and communication (ISAC) network empowered by a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS). This architecture offers seamless 360° coverage, supporting the coexistence of aerial and ground users in the low-altitude economy, while leveraging near-field propagation to enhance both communication and sensing performance. However, the heterogeneity of users and the coupling between sensing, communication, and energy consumption pose challenges for efficient beamforming. To this end, we propose a joint beamforming and phase shift optimization framework to minimize average power consumption under communication delay and sensing constraints. To cope with the long-term characteristics, we design a Lyapunov-based deep deterministic policy gradient (LyDDPG) algorithm to transform long-term objectives into slot-level decisions, achieving stable performance in dynamic environments. To further improve constraint satisfaction and policy convergence, we further propose a robust policy deep reinforcement learning (RPDRL) algorithm, which integrates policy-based reinforcement learning with successive convex approximation. The Lyapunov transformation is embedded into the RPDRL to dynamically adjust weighting factors, accelerating policy learning. Simulation results demonstrate the effectiveness and superiority of the proposed methods in achieving energy-efficient, constraint-aware beamforming in near-field ISAC networks.